Appen Limited — AI supply-chain exposure

Appen is an Australian-listed training-data company that gathers, labels, and quality-checks the human-annotated data model developers use to teach and refine large language models. Its work spans annotation, reinforcement learning from human feedback, and evaluation of model outputs. It sits at the raw-input edge of the chain, upstream of the models themselves — a supplier of pretraining and alignment data rather than a builder of models, an early leader in the field repositioned for the LLM era.

Its exposure is to the labour-intensive, human-in-the-loop layer that every lab depends on but few want to run in-house. That work is recurring, but it is also contestable — labs can bring annotation in-house or lean on synthetic data, and the business carries few switching costs. The model places it near the periphery: structurally connected to frontier training, but with thin pricing power because what it sells is a service, not a scarce asset.